RE: [PATCH v4 3/8] mm: Add support for PUD-sized transparent hugepages

[Wilcox, Matthew R]

It's awful.  I have a v6 in the works which fixes a number of problems in v5, but there are about ten Kconfig options which the x86 code depends on.  And then ... yes, there's this bad definition of pud_t on ARM.  Arnd has a patch to fix that problem, Subject: [PATCH v2] [RFC] ARM: modify pgd_t definition for TRANSPARENT_HUGEPAGE_PUD.

But 0day is still pointing out other problems with the current patchset, so I'd hold off on it until I get v6 posted if I were you.  Thanks for picking up x86-unify-native__get_and_clear-smp-case.patch

-----Original Message-----
From: Andrew Morton [mailto:akpm@xxxxxxxxxxxxxxxxxxxx] 
Sent: Tuesday, March 29, 2016 3:17 PM
To: Wilcox, Matthew R
Cc: Matthew Wilcox; linux-mm@xxxxxxxxx; linux-nvdimm@xxxxxxxxxxx; linux-fsdevel@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; x86@xxxxxxxxxx
Subject: Re: [PATCH v4 3/8] mm: Add support for PUD-sized transparent hugepages

On Sun, 31 Jan 2016 23:09:30 +1100 Matthew Wilcox <matthew.r.wilcox@xxxxxxxxx> wrote:

> From: Matthew Wilcox <willy@xxxxxxxxxxxxxxx>
> 
> The current transparent hugepage code only supports PMDs.  This patch
> adds support for transparent use of PUDs with DAX.  It does not include
> support for anonymous pages.
> 
> Most of this patch simply parallels the work that was done for huge PMDs.
> The only major difference is how the new ->pud_entry method in mm_walk
> works.  The ->pmd_entry method replaces the ->pte_entry method, whereas
> the ->pud_entry method works along with either ->pmd_entry or ->pte_entry.
> The pagewalk code takes care of locking the PUD before calling ->pud_walk,
> so handlers do not need to worry whether the PUD is stable.

Why is this patchset always so hard to compile :(

> ...
>
> --- a/include/linux/pfn_t.h
> +++ b/include/linux/pfn_t.h
> @@ -82,6 +82,13 @@ static inline pmd_t pfn_t_pmd(pfn_t pfn, pgprot_t pgprot)
>  {
>  	return pfn_pmd(pfn_t_to_pfn(pfn), pgprot);
>  }
> +
> +#ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
> +static inline pud_t pfn_t_pud(pfn_t pfn, pgprot_t pgprot)
> +{
> +	return pfn_pud(pfn_t_to_pfn(pfn), pgprot);
> +}
> +#endif
>  #endif
>  
>  #ifdef __HAVE_ARCH_PTE_DEVMAP
> @@ -98,5 +105,6 @@ static inline bool pfn_t_devmap(pfn_t pfn)
>  }
>  pte_t pte_mkdevmap(pte_t pte);
>  pmd_t pmd_mkdevmap(pmd_t pmd);
> +pud_t pud_mkdevmap(pud_t pud);

arm allnoconfig:

In file included from kernel/memremap.c:17:
include/linux/pfn_t.h:107: error: 'pud_mkdevmap' declared as function returning an array
because it expands to

pgd_t pud_mkdevmap(pgd_t pud);

and

typedef unsigned long pgd_t[2];                                                 


Also the patch provides no implementation of pud_mkdevmap() so it's
obviously going to break bisection.